During sentence processing, comprehenders form different predictions regarding the unfolding of the sentence. Every act of predictive structure building puts the parser in greater risk of having to perform a costly reanalysis. The current study investigates how the parser deals with evidence against a predicted structural dependency, and asks what determines the difficulty of the reanalysis. I suggest that (at least) three different factors affect the reanalysis process: the motivation for the prediction, the disparity between the initial and the new reading, the type of cue that indicates the need for reanalysis. In a series of psycholinguistic experiments, focusing on the predictive processing of filler-gap dependencies, I present evidence for the effect of these factors on incremental formation and transformation of sentential representations.
